Lab: Semantic Lifting - CSV

From info216
Revision as of 17:51, 18 February 2023 by Sinoa (talk | contribs) (→‎Tasks)

Topic

  • Reading non-semantic data tables into semantic knowledge graphs
  • Specifically, reading data in CSV format via Pandas dataframes into RDF graphs

Useful materials

The textbook (Allemang, Hendler & Gandon):

Pandas:

rdflib:

  • classes/interfaces from earlier (such as Graph, Namespace, URIRef, Literal, perhaps BNode)
  • also vocabulary classes like RDF (e.g., type), RDFS (e.g., label) and XSD (for various datatypes)

Tasks

We will be working with the same dataset as in the SPARQL exercise: FiveThirtyEight's Russia Investigation. It contains data about special investigations conducted by the United States from the Watergate-investigation until May 2017. This page explains the Russia Investigation dataset a bit more.

Task: In the SPARQL exercise, you downloaded the data as a Turtle file (File:Russia investigation kg.txt, which you renamed to .ttl). This time you will download the data as a CSV file from GitHub.

(Unfortunately, the earlier exercises used a wrong and very unfinished version of the russia_investigation_kg.txt file. It has been updated now with the correct version!)

Task: Install Pandas in your virtual environment, for example

pip install pandas

Write a Python program that imports the pandas API and uses Pandas' read_csv function to load the russia-investigation.csv dataset into a Pandas dataframe.

Task: (Pandas basics) Inspect the Pandas dataframe. If you have called your dataframe df, you can check out the expressions below. Use the documentation to understand what each of them does.

df.shape
df.index  # ...and list(df.index)
df.columns
df['name']
df.name
df.loc[3]
df.loc[3]['president']

(Pandas offers many ways of picking out rows, columns, and values. These are just examples to get started.)

Task: (Pandas basics) Pandas' apply method offers a compact way to process all the rows in a dataframe. This line lists all the rows in your dataframe as a Python dict():

df.apply(lambda row: print(dict(row)), axis=1)

What happens if you drop the axis argument, or set axis=0?

Task: Instead of the lambda function, you can use a named function. Write a function that prints out only the name and indictment-days in a row, and use it to print out the name and indictment-days for all rows in the dataframe.

Alternative to df.apply(): Pandas offers several ways to iterate through data. You can also use the itertuples methods in a simple for-loop to iterate through rows.

Task: Modify your function so it adds name and indictment-days triples to a global rdflib Graph for each row in the dataframe. The subject in each triple could be the numeric index of the row.

You can use standard terms from RDF, RDFS, XSD, and other vocabularies when you see fit. Otherwise, just use an example-prefix.

Things may be easier if you copy df.index into an ordinary column of the dataframe:

df['id'] = df.index

You can use this index, along with a prefix, as the subject in your triples.

Task: Continue to extend your function to convert the non-semantic CSV dataset into a semantic RDF one. Here is an example of how the data for one investigation could look like in the end:

muellerkg:investigation_0 a muellerkg:Indictment ;
    muellerkg:american true ;
    muellerkg:cp_date "1973-01-30"^^xsd:date ;
    muellerkg:cp_days -109 ;
    muellerkg:indictment_days -246 ;
    muellerkg:investigation muellerkg:watergate ;
    muellerkg:investigation_days 1492 ;
    muellerkg:investigation_end "1977-06-19"^^xsd:date ;
    muellerkg:investigation_start "1973-05-19"^^xsd:date ;
    muellerkg:name muellerkg:James_W._McCord ;
    muellerkg:outcome muellerkg:conviction ;
    muellerkg:overturned false ;
    muellerkg:pardoned false ;
    muellerkg:president muellerkg:Richard_Nixon .

If you have more time

Task: If you have not done so already, you should include checks to ensure that you do not add empty columns to your graph.

Task: If you have more time, you can use DBpedia Spotlight to try to link the people (and other "named entities") mentioned in the dataset to DBpedia resources.

pip install pyspotlight

You can start with the code example below, but you will need exception-handling when DBpedia is unable to find a match. For instance:

import spotlight

ENDPOINT = 'https://api.dbpedia-spotlight.org/en/annotate'
CONFIDENCE = 0.5  # filter out results with lower confidence

def annotate_entity(entity_name, filters={'types': 'DBpedia:Person'}):
    annotations = []
    try:
	annotations = spotlight.annotate(ENDPOINT, entity_name, confidence=CONFIDENCE, filters=filters)
    except spotlight.SpotlightException as e:
        # catch exceptions thrown from Spotlight, for example when no DBpedia resource is found
	print(e)
	# handle exceptions here
    return annotations

The example uses the types-filter with DBpedia:Person, because we only want it to match with people. You can choose to only implement the URIs in the response, or the types as well.

Useful materials: